---

NEW (June 2021)

• This branch has code for Pytorch 0.4 and Python 2. Please use the code in this branch titled Python3 of this repository for code that works with Python 3 and Pytorch 1.4.0
• The branch also has new links to download trained checkpoints. (the older link no longer works)

---

This folder contains code to replicate results corresponding to B, BC, BCP in Table 1 and B-R5, BC-R5, BCP-R5 in Table 3 in the paper.

Instructions for use

• Please download the contents of this folder first. Deeplab v2 authors released a pretrained version of their net on MSCOCO. We converted it to a pytorch .pth file and use it to fine train our model. The downloaded folder contains the file MS_DeepLab_resnet_pretained_VOC.pth which corresponds to this .pth file. Please keep this .pth file in the same folder as the train script. The folder also contains train_sketches.zip and train_sketch_GT.zip which contains the augmented data (sketches and the corresponding ground truth respectively) used to train our model.

• Use train_r5.py to train model for results B-R5, BC-R5, BCP-R5 in Table 3.

For BCP-R5, run

python train_r5.py --segnetLoss --lambda1 1.0 --GTpath <train gt images path here> --IMpath <train images path here> 

For BC-R5, run

python train_r5.py --segnetLoss --lambda1 0.0 --GTpath <train gt path here> --IMpath <train images path here> 

For B-R5, run

python train_r5.py --lambda1 0.0 --GTpath <train gt images path here> --IMpath <train images path here> 

To get a description for each flag used, run

python train_r5.py -h

• To evaluate models trained with train_r5.py, run eval_r5.py with the required snapPrefix, testGTpath(test ground truth) and testIMpath(test images) arguments. Links to download the test data is available in this readme.

• Use train_r1.py to train model for results B, BC, BCP in Table 1.

For BCP, run

python train_r1.py --segnetLoss --lambda1 1.0 --GTpath <train gt images path here> --IMpath <train images path here> 

To get a description for each flag used, run

python train_r1.py -h

• To evaluate models trained using train_r1.py, run eval_r1.py with the required snapPrefix, testGTpath(test ground truth) and testIMpath(test images) arguments. Links to download the test data is available in this readme

• Run table1.py, table3.py to evaluate the downloaded .pth files and get results corresponding to table 1 and table 3 in the paper. Download pretrained models from here and place them in the folder data/snapshots. Download the annotated sketches(using instructions in this readme) and place them in 'data/sketch-dataset'. After this, table1.py and table3.py should run properly.

pred_gt.txt contains the predicted label and the real label of each sketch image. The predicted label can also be found real time using a trained classifier. The architecture of the classifier is mentioned in the supplementary material and the training policy / data is mentioned in the paper. This file is used during evaluation.